[OpenAFS] Moving Volumes

[Michael Aldrich]

> Let's say that it takes 2 minutes to move a volume.  There will be an
> outage for some smaller interval while the move happens.
>
> By the way, should you be using RO volumes for your web servers?  That
> keeps them available even when your RW is unavailable.

I do have RO sites for every volume in my cell.

> The way moving a RW volume works is this.  (a) The RW volume is cloned
> to a temporary clone volume.  (b) This temporary clone volume is then
> shipped (via volume dump and volume restore) to the destination site.
> (c) An incremental dump of the RW volume is shipped to an incremental
> restore on the destination site.  (d) The source volume is deleted as
> all location sites are told where the volume now lives.  I'm ignoring
> some details here, but they don't matter for your purposes.

The volumes are rarely modified by staff. They contain static pages for my
site. If the RW volume is being moved, can the RO volume take its place in
order to ensure no outage?

> Access to the RW volume is suspended selectively during parts of the
> above, and that's when your web servers might potentially be suspended
> as well; they will of course wait without the filesystem showing
> anything other than a delay (their system calls will complete normally).
>  So during step (a), access to the RW is suspended, but that's only a
> few seconds.  During step (b), access is *permitted*, and this is likely
> to be the longest phase of the operation--as all files in the RW volume
> are shipped across a network.  During step (c), access to the RW is
> again suspended, but only changed files are shipped--ones that changed
> during the (b) phase.  So if nothing changes during the (b) phase, step
> (c) will execute quickly.  Step (d) takes at most a few seconds.
>
> Craig
> _______________________________________________
> OpenAFS-info mailing list
> OpenAFS-info@openafs.org
> https://lists.openafs.org/mailman/listinfo/openafs-info
>